Two-branch network architecture has shown its efficiency and effectiveness in real-time semantic segmentation tasks. However, direct fusion of high-resolution details and low-frequency context has the drawback of detailed features being easily overwhelmed by surrounding contextual information. This overshoot phenomenon limits the improvement of the segmentation accuracy of existing two-branch models. In this paper, we make a connection between Convolutional Neural Networks (CNN) and Proportional-Integral-Derivative (PID) controllers and reveal that a two-branch network is equivalent to a Proportional-Integral (PI) controller, which inherently suffers from similar overshoot issues. To alleviate this problem, we propose a novel three-branch network architecture: PIDNet, which contains three branches to parse detailed, context and boundary information, respectively, and employs boundary attention to guide the fusion of detailed and context branches. Our family of PIDNets achieve the best trade-off between inference speed and accuracy and their accuracy surpasses all the existing models with similar inference speed on the Cityscapes and CamVid datasets. Specifically, PIDNet-S achieves 78.6% mIOU with inference speed of 93.2 FPS on Cityscapes and 80.1% mIOU with speed of 153.7 FPS on CamVid.

In this paper we derive some new and useful conditions on the open loop transfer function, necessary for closed loop stability. The derivation is based on imposing the sign-invariance of the closed loop characteristic polynomial evaluated on the real axis. Although this is a simple necessary condition for closed loop stability, it leads to fairly stringent requirements on the open loop transfer function. Also, the results can be stated only in terms of the numbers of real poles and zeros of the open loop system even though the system may have complex poles and zeros. Using the same approach new necessary conditions for stabilizability by P, PI and PID controllers as well as by arbitrary controllers are also presented.

In this paper we develop a systematic approach to the design of a digital controller for a linear plant. The digital controller is parametrized by an unknown or design, real parameter vector k, which may represent for instance, the coefficients of the digital controller transfer function. The stabilizing set ${\mathcal{S}}$ in the space of these design parameters is an important and crucial piece of information which is needed to initiate the design process, since every design must reside in this set. In particular multiobjective design is impossible without determining this set. In general this set is not convex or even connected, its computation is di cult and few results are available. In this paper, we use Mikhailov’s criterion and the Tchebychev representation of a polynomial, evaluated on the unit circle, to determine an inner approximation ${{\mathcal{S}}_i}$ of ${\mathcal{S}}$, which is convex and described by linear inequalities. Applications of this result to the design of controllers are presented. In particular we show that a geometric approach to gain and phase margin based designs are feasible in this framework. Illustrative examples are presented with a focus on the practically important case of PI and PID controllers. The results should have application to many areas of digital controller design such as hard disk servomechanisms, driverless cars and robots.

Integrity refers to the property that a system remains functional despite failures of certain subsystems. In multivariable systems controller integrity means that stability is preserved despite failures of certain sensors or actuators. In this paper we first show by examples that typically multivariable controllers designed by the standard state feedback observer approach lack integrity even when plant and controller are open loop stable. With this background as motivation we develop a technique of designing controllers that do possess integrity. The main mathematical tools are our recent results on computation of sets of stabilizing PID controllers and the Mapping Theorem of Robust Control Theory. The method can be extended beyond PID control and also to the plant integrity problem. Illustrative examples are included and areas for further research are indicated.

A complete stabilizing PID set is developed without the need for an identified analytical model, for an automotive cruise-control system from its frequency response. The resulting set serves as a framework and is used to determine the PID gains satisfying certain performance indices. These performance measures are derived from the need for efficient tracking and disturbance rejection. This ensures that the controller is tuned with the most appropriate PID gains, yielding the desired closed-loop response. Also, this serves as a viable alternative to the iterative method of estimating stabilizing PID gains.

In this paper, we present new results that are useful for the synthesis of fixed order digital controllers from plant frequency response data. The main result is a nonparametric characterization of the frequency response of arbitrary order rational discrete time stabilizing controllers. This result shows that the frequency response of the inverse plant imposes constraints on the magnitude, phase, and rate of change of phase of the digital controller's frequency response at certain frequencies.

In this paper, we show how a two-parameter compensator can always be designed for any Linear Time Invariant (LTI) plant, that does not have a zero at the origin, to render its step response non-overshooting.

P. Bondi et al. (IEEE Journal of Robotics and Autom., vol.4, p.14-22, Feb. 1988) proposed a high-gain feedback method for learning control and showed that there exist gains which ensure trajectory convergence to a desired response. An adaptive gain-adjustment technique is suggested that enables the system to choose adaptively the gains which yield convergence. The technique is illustrated with the results of a simulation.>